Random Coefficient Transfer Function Model

• Activity Number: 231
• Type: Contributed
• Date/Time: Tuesday, August 9, 2005 : 8:30 AM to 10:20 AM
• Sponsor: Business and Economics Statistics Section
• Author(s): Hyunyoung Choi*+ and Jeff Douglas and Bonnie Ray
• Companies: University of Illinois, Urbana Champaign and University of Illinois, Urbana Champaign and IBM
• Keywords: Random Transfer Function Model ; Random Intervention Model ; Random Coefficient Model ; Gibbs Sampling ; Nonlinear Optimization ; Maximum Likelihood Estimation

Abstract

A transfer function model describes how the level of input influences the level of the system output. Identification of transfer function model consists of identification of transfer function and the baseline model. In panel data analysis, it is more natural to assume random effect for the parameters in transfer function and baseline. By allowing transition period and the random ramp up effect for the transfer function, the model becomes nonlinear and the posterior distribution become more complicated with unknown form. We propose linear/nonlinear transfer function panel model with random effect and adapt the Gibbs sampler to implement the Bayesian paradigm. It also is common to observe missing data, and Gibbs sampler approach gives a straightforward solution to the distribution problem due to missing data. Using simulation, we compare the estimation from univariate transfer function model and panel transfer function model with random effect. We then present an application to the impact on stock returns due to top management change for goals such as estimation of the baseline and the impact of stimulus.